Give up on protecting against segfaults in aq_* functions, when we encounter __parent__ circles with more than one intermediate object.

Raise `RuntimeError: Recursion detected in acquisition wrapper` if an object with a `__parent__` pointer points to a wrapper that in turn points to the original object.

Fix the triple parent test - we actually get reasonable results now, but also some infinite recursion on non-existant attributes

Avoid an infinite recursion in parent parent circles, if they are no longer wrapped in all cases like with ExtensionClass 4.0

Fixed a segfault on 64bit platforms when providing the `explicit` argument to the aq_acquire method of an Acquisition wrapper. Thx to LP #675064 for the hint to the solution. The code passed an int instead of a pointer into a function.

  ``unicode(wrapped)`` was calling ``__str__`` with an unwrapped ``self``.

(reported by David Glick on zope-dev)

Document strange behavior with parent pointer cycles involving an intermediate object, but show that these don't lead to infinite recursion

Give both wrapper classes a ``__getnewargs__`` method, which causes the ZODB optimization to fail and create persistent references using the ``_p_oid`` alone. This happens to be the persistent oid of the wrapped object. This lets these objects to be persisted correctly, even though they are passed to the ZODB in a wrapped state.

Added failing tests for http://dev.plone.org/plone/ticket/10318. This shows an edge-case where AQ wrappers can be pickled using the specific combination of cPickle, pickle protocol one and a custom Pickler class with an ``inst_persistent_id`` hook. Unfortunately this is the exact combination used by ZODB3.